Open-source guardrail

The agent editing its standing instructions

AgentTrail Guard asks you about this by default: you decide before the call runs. Where an agent cannot ask, the call is refused instead.

Default action
Ask
Severity
High severity
Library version
0.2.1Sep 29, 2026

What it catches, and what it misses

Written into the rule itself, next to what it matches, so you can judge it before you trust it.

Holds a file tool opening the instructions a coding agent loads at the start of every session: CLAUDE.md, CLAUDE.local.md and .claude/rules/; AGENTS.md and AGENTS.override.md, which Codex, Cursor, Windsurf, Copilot and Cline all read; GEMINI.md; .cursorrules; Windsurf's .windsurfrules, .windsurf/rules/ and .devin/rules/; Cline's .clinerules file or directory and its global Cline/Rules/ folder; Copilot's .github/copilot-instructions.md and .github/instructions/**/*.instructions.md; and .aider.conf.yml, which sets the files Aider reads on every launch. A line written into one of these is followed in every later session. Matched in any directory and in any letter case. Reading is excluded — the Read and Grep tools never match — so an agent opening its instructions to read them is not held; every other file tool is, including one this corpus does not know. File tools carry a path and no content, so it still cannot see what was written. Does not cover .cursor/rules/, which fs.agent-self-config holds. Misses a context file renamed through Gemini's context.fileName or Codex's project_doc_fallback_filenames, Aider's CONVENTIONS.md, which Aider loads only when asked and which is too common a name to match, and any of these files written by a shell command instead of a file tool.

Tested on every build

The rule must match every command on the left and none on the right, or the library does not build. Catching the real thing is easy; staying quiet on the near-miss is the hard part.

Catches (18)

  • EditCLAUDE.md
  • Writepackages/api/CLAUDE.md
  • EditCLAUDE.local.md
  • Write.claude/rules/testing.md
  • EditAGENTS.md
  • Editservices/billing/AGENTS.override.md
  • Write/home/dev/.gemini/GEMINI.md
  • Edit.cursorrules
  • Edit.windsurfrules
  • Edit.windsurf/rules/style.md
  • Edit.devin/rules/style.md
  • Edit.clinerules
  • Write.clinerules/coding.md
  • Write/Users/dev/Documents/Cline/Rules/global.md
  • Edit.github/copilot-instructions.md
  • Edit.github/instructions/frontend/react.instructions.md
  • Edit.aider.conf.yml
  • Editdocs/claude.md

Stays quiet on (13)

  • EditREADME.md
  • EditCLAUDE.md.bak
  • EditCONVENTIONS.md
  • Editdocs/agents/overview.md
  • Editsrc/agents.ts
  • Editdocs/rules/style.md
  • Edit.github/PULL_REQUEST_TEMPLATE.md
  • Edit.github/workflows/ci.yml
  • Edit.cursor/rules/style.mdc
  • Edit.claude/settings.json
  • Edit.aider.chat.history.md
  • ReadCLAUDE.md
  • Grep.github/copilot-instructions.md

What "ask" means in each app

The guard runs as a hook in each app, and each app gives a hook different powers. Here is what this rule's default action does in each one.

Claude Code
You are asked before the call runs. If your settings.json already allows the tool outright, Claude Code runs it with no prompt.
Cursor
A terminal command that reaches the guard gets Cursor's approval card. Any other call that needs approval, such as a file edit or a read, is refused, because Cursor cannot ask there.
Codex CLI
Codex cannot ask, so the call is refused, and the reason says a person has to approve it.

This rule matches file paths. In Codex CLI, reading a file fires no hook, and in Cursor, reads shown as Explored are not checked, so there it sees fewer calls than in Claude Code. Each app hands the guard a different set of calls: in a Cursor sandbox run mode, for one, some terminal commands run without reaching the guard at all. Read the notes for Cursor and for Codex CLI before you rely on a rule there.

Change what it does in one command

Turn it off, change its action, or silence it on one command shape. The narrow one is allow: the rule keeps catching everything else.

agenttrail-guard
$agenttrail-guard guardrails show ac.instruction-file-edit# everything about it
$agenttrail-guard guardrails set-action ac.instruction-file-edit warn# block, ask or warn
$agenttrail-guard guardrails allow ac.instruction-file-edit '<pattern>'# silence one shape
$agenttrail-guard guardrails disable ac.instruction-file-edit# turn it off

Put these guardrails in front of your agent.

AgentTrail Guard is free and open source. It checks every command and file change against the whole library before your agent runs it, on your machine, with no account.

bash
$npm i -g @agenttrail/guard
Read the source on GitHub